A method and device for generating technical mark

By combining a large text generation model and a resource library, high-quality technical bids are generated, solving the problems of low efficiency and limitations in existing technical bid generation, and achieving efficient and personalized bid generation.

CN118569222BActive Publication Date: 2024-11-22SHANGHAI YUZHIJING INTELLIGENT TECH CO LTD
View PDF 5 Cites 0 Cited by

Patent Information

Application Number
CN202411035211.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-07-31
Publication Date
2024-11-22
Estimated Expiration
2044-07-31

AI Technical Summary

Technical Problem

Existing methods for generating technical tenders are inefficient, fail to fully understand user needs, resulting in incomplete and inefficient tender documents. Furthermore, they are limited by the availability of material libraries and are subject to outdated technology.

Method used

The method combines text-generated large models and material libraries. The data preparation unit collects project overview and scoring details, the title generation unit generates and scores technical title, and the technical title content generation unit merges the content from the material library and the large model to generate high-quality technical titles.

Benefits of technology

It improves the efficiency and quality of technical bid generation, reduces labor costs, and generates bid documents that better meet user needs, reduce redundant information, and adapt to the needs of different types of bid documents.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118569222B_ABST
    Figure CN118569222B_ABST
Patent Text Reader

Abstract

The present invention discloses a method and device for generating a technical mark, and relates to the technical field of electronic digital data processing; the device includes a data preparation unit, a title generation unit and a technical mark content generation unit, and the method includes the steps of S1 data preparation, S2 title generation and S3 technical mark content generation, S2 title generation, generating the title of the technical mark and scoring the title, repeatedly generating the title of the technical mark and scoring the title, and selecting the title with the highest score as the final title; S3 technical mark content generation, grouping to obtain a material library title group and a large model title group, and obtaining corresponding content from the material library according to the material library title; according to the large model title, generating corresponding content through text to generate a large model, and merging the content corresponding to the material library title group and the content corresponding to the large model title group through the index when grouping to obtain the technical mark. It generates technical marks with high efficiency through the steps of data preparation, title generation and technical mark content generation, etc.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of electronic digital data processing, and in particular to a method and device for generating a technical mark. Background Art

[0002] Technical specifications refer to the technical contents in the bidding documents. They are the responsiveness to the substantive requirements and conditions put forward in the bidding documents, the construction organization design or plan, and the commitment to the special technical contents in the bidding documents. Technical specifications play an important role in the entire bidding document.

[0003] At present, the market for tender documents based on AI technology is not yet mature. There are two main related technologies on the market:

[0004] 1. Generate a paragraph of explanatory text through a simple question-and-answer method to provide the customer with some tender information.

[0005] 2. Collect historical technical bids or purchase finished technical bids as a material library, and generate bid documents based on the material library.

[0006] There are obvious defects in the two current methods of writing bids as follows:

[0007] Technology 1: Understanding user needs only through dialogue and providing partial feedback on the bid document cannot allow customers to automatically generate a technical bid. Customers still need to spend time writing the bid document, and it is impossible to fully understand user needs only through dialogue. Therefore, this method has three disadvantages: 1. Ineffective information acquisition, wasting user time; 2. Low efficiency, requiring users to have cyclical dialogues; 3. Unable to realize the automatic generation of bid documents, and the commercial implementation of automated bid generation has not been truly realized.

[0008] Technique 2: Only use the material library to find similar bids, and then give them to customers or use them as templates for users to imitate. There are two main disadvantages: 1. The collected material library is limited. When the material library is exhausted, the generation of bids will be restricted; 2. Relying solely on existing materials leads to limitations in the generation of bids, which will cause bids pieced together with older materials to be different from the needs of the times. Summary of the invention

[0009] The present invention provides a method and device for generating a technical mark, which solves the technical problem of low efficiency in generating a technical mark.

[0010] In order to solve the above technical problems, the technical solutions adopted by the present invention are as follows:

[0011] A device for generating a technical bid includes a data preparation unit, a title generation unit and a technical bid content generation unit. The data preparation unit is used to collect and obtain a user's project profile, project theme and project scoring rules, and generate a material library based on historical technical bid documents; the title generation unit is used to input the project profile and project theme as keywords into a text generation model based on the project profile, project theme and scoring rules obtained by the data preparation unit, and generate a technical bid title; the text generation model reads the scoring rules and project theme input by the user and the added industry experience or general requirements for the title, and scores the title of the generated technical bid; the title of the technical bid and the title scoring are repeated, and the best one is selected. The title corresponding to the high score is taken as the final title; a technical mark content generation unit is used to obtain multiple final titles according to the title generation unit, group the multiple final titles to obtain a material library title group and a large model title group, select from the material library according to each material library title in the material library title group to obtain the content corresponding to the material library title and obtain the content corresponding to the material library title group; according to each large model title in the large model title group, fill the project overview and the large model title as keywords into the prompt word, generate the content corresponding to the large model title through the text and obtain the content corresponding to the large model title group, merge the content corresponding to the material library title group and the content corresponding to the large model title group through the index during grouping, and generate the technical mark.

[0012] A further technical solution is that: the data preparation unit is also used to obtain the project theme by giving users optional options, and the user's needs are used as the direction for generating the content of the bid document. The material library is a file with classification, title, summary and technical specification content, and the material library is a docx or doc file; the content of each technical specification is marked with [start] as the beginning and [end] as the end, and each technical specification is distinguished by number.

[0013] A further technical solution is that the title generation unit is also used to repeatedly generate the title of the technical mark and the title score 3 to 5 times.

[0014] A method for generating a technical bid, used in the above-mentioned device for generating a technical bid, includes the steps of S1 data preparation, S2 title generation and S3 technical bid content generation, S1 data preparation, collecting and obtaining the user's project overview, project theme and project scoring criteria, and generating a material library based on historical technical bid documents; S2 title generation, according to the project overview, project theme and scoring criteria obtained by the data preparation unit, inputting the project overview and project theme as keywords into a text generation model to generate a technical bid title; the text generation model reads the scoring criteria and project theme input by the user and the added industry experience or general requirements for the title, and scores the title of the generated technical bid; repeatedly generating the technical bid title and Title scoring, select the title corresponding to the highest score as the final title; S3 technical standard content generation, obtain multiple final titles according to the title generation unit, group the multiple final titles to obtain the material library title group and the large model title group, select from the material library according to each material library title in the material library title group to obtain the content corresponding to the material library title and obtain the content corresponding to the material library title group; according to each large model title in the large model title group, fill the project overview and the large model title as keywords into the prompt word, generate the content corresponding to the large model title through the text and obtain the content corresponding to the large model title group, merge the content corresponding to the material library title group and the content corresponding to the large model title group through the index during grouping to generate the technical standard.

[0015] The beneficial effects of adopting the above technical solution are:

[0016] A device for generating a technical bid includes a data preparation unit, a title generation unit and a technical bid content generation unit. The data preparation unit is used to collect and obtain a user's project profile, project theme and project scoring rules, and generate a material library based on historical technical bid documents; the title generation unit is used to input the project profile and project theme as keywords into a text generation model based on the project profile, project theme and scoring rules obtained by the data preparation unit, and generate a technical bid title; the text generation model reads the scoring rules and project theme input by the user and the added industry experience or general requirements for the title, and scores the title of the generated technical bid; the title of the technical bid and the title scoring are repeated, and the best one is selected. The title corresponding to the high score is used as the final title; the technical mark content generation unit is used to obtain multiple final titles according to the title generation unit, group the multiple final titles to obtain the material library title group and the large model title group, select from the material library according to each material library title in the material library title group to obtain the content corresponding to the material library title and obtain the content corresponding to the material library title group; according to each large model title in the large model title group, fill the project overview and the large model title as keywords into the prompt word, generate the large model through the text to generate the content corresponding to the large model title and obtain the content corresponding to the large model title group, merge the content corresponding to the material library title group and the content corresponding to the large model title group through the index during grouping, and generate the technical mark. It generates the technical mark with high efficiency through the data preparation unit, the title generation unit and the technical mark content generation unit.

[0017] A method for generating a technical bid, used in the above-mentioned device for generating a technical bid, includes the steps of S1 data preparation, S2 title generation and S3 technical bid content generation, S1 data preparation, collecting and obtaining the user's project overview, project theme and project scoring criteria, and generating a material library based on historical technical bid documents; S2 title generation, according to the project overview, project theme and scoring criteria obtained by the data preparation unit, inputting the project overview and project theme as keywords into a text generation model to generate a technical bid title; the text generation model reads the scoring criteria and project theme input by the user and the added industry experience or general requirements for the title, and scores the title of the generated technical bid; repeatedly generating the technical bid title and Title scoring, select the title corresponding to the highest score as the final title; S3 technical mark content generation, obtain multiple final titles according to the title generation unit, group the multiple final titles to obtain the material library title group and the large model title group, select from the material library according to each material library title in the material library title group to obtain the content corresponding to the material library title and obtain the content corresponding to the material library title group; according to each large model title in the large model title group, fill the project overview and the large model title as keywords into the prompt word, generate the large model through the text to generate the content corresponding to the large model title and obtain the content corresponding to the large model title group, merge the content corresponding to the material library title group and the content corresponding to the large model title group through the index during grouping, and generate the technical mark. It generates technical marks efficiently through the steps of data preparation, title generation and technical mark content generation.

[0018] Please refer to the detailed description of the specific implementation method. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 is a principle block diagram of embodiment 1 of the present invention;

[0020] Figure 2 is a flow chart of the data preparation steps in Embodiment 2 of the present invention;

[0021] Figure 3 is a flowchart of the title generation step in Embodiment 2 of the present invention;

[0022] Figure 4 It is a screenshot composed of files in the library;

[0023] Figure 5 It is a flowchart of the steps of generating technical specification content in Embodiment 2 of the present invention. DETAILED DESCRIPTION

[0024] The shortcomings of similar methods on the market are: 1. This method guides users to generate technical bids through a historical technical bid material library, but it cannot generate a complete bid; 2. This method generates bids through dialogue with users, but the content generated by this method is inaccurate and wastes user time.

[0025] This application aims to solve the above technical problems and achieve high-quality automatic generation of technical targets.

[0026] The present application provides a solution for fully automatically generating technical bids in bids by combining a text-generated large model and a material library, wherein the large model is GPT, Baichuan, Qianwen or ChatGLM. It solves the technical problem that technical bids cannot be generated fully automatically and that the generation process is overly dependent on the material library. Different scoring criteria are formulated according to different types of bids. In the large model question and answer, the question and answer results are scored in real time to generate high-quality technical bid content. Users can use the technical solution of this application to upload project information, use specific scoring criteria for different types of bids, or customize scoring criteria to obtain a completed presentation document, lower the threshold for writing technical bids, and reduce labor costs and time consumption.

[0027] Technical contributions of this application:

[0028] 1. The technical solution of this application is based on the project profile uploaded by the user. Under the guidance of the scoring rules, better prompt words are written through industry experience and the requirements entered by the user to optimize the logic of generating technical bids, so as to obtain a high-quality technical bid that meets user needs.

[0029] The high quality means that the content and structure are more in line with the requirements of the technical specifications as a whole, and the compliance with user needs means that users can guide the generation of technical specifications according to their own scoring criteria.

[0030] 2. The technical solution of this application uses a large text generation model combined with a material library method to generate technical bids. At the same time, the large model of the text generation model and the material library method can improve the diversity and quality of the generated content, reduce repetition, and reduce redundant information, making the final generated bid more comprehensive and rich. The combined use of the two methods can improve the efficiency of bid generation and save time and labor costs. The content generated by the text generation model can meet the personalized needs of users, while the content in the material library provides existing excellent practices and standardized content, making the generation process more efficient and convenient.

[0031] The text generation model is GPT, Baichuan, Qianwen or ChatGLM.

[0032] 3. The technical solution of this application uses a large text generation model to generate technical bid titles, and scores them, and selects high-scoring titles as the final technical bid titles. Traditionally, evaluating titles usually requires relying on manual subjective judgment. This method uses a large text generation model to evaluate the score of the title, which makes the evaluation process more objective, fast and efficient. The large text generation model can quantitatively evaluate the quality of the title based on existing data and knowledge, thereby providing an objective basis for the final selection and improving the accuracy and consistency of the evaluation.

[0033] 4. The technical solution of this application proposes a device for automatically generating technical bids, a method for generating complete technical bids based on limited material resources. This technology includes algorithms and programs that combine text generation models with material libraries to automatically and efficiently generate technical bid content.

[0034] Compared with the existing technical solutions, the technical solution of this application has the following advantages:

[0035] 1. The technical solution of this application generates a complete technical specification based on limited material resources.

[0036] 2. The technical solution of the present application reduces the negative impact caused by the time limit and field mismatch caused by the material library.

[0037] 3. Based on user needs, according to the project profile uploaded by the user and under the guidance of the scoring criteria, optimize the logic of generating bids and generate a high-quality technical bid that meets user needs.

[0038] 4. Automatically generating technical bids can greatly reduce the time and labor costs required to write technical bids. Through the combination of the question-and-answer big model and the material library, the technical solution of this application can quickly generate technical bid content that meets the requirements of the bid document and improve work efficiency.

[0039] 5. Users can choose different types of bids and scoring criteria according to the characteristics of the project, or they can customize the scoring criteria to meet the needs of different projects, making the generated technical bids more flexible and personalized.

[0040] 6. The technical solution of this application lowers the threshold for writing technical bids. Even personnel without professional technical background can easily write high-quality technical bids, thereby promoting the popularization and standardization of bid writing.

[0041] Big model: refers to a large neural network model that can generate various types of content. These models are trained to understand and generate data in various forms such as text, images, audio, etc. Big models are usually based on deep learning architectures, such as transformers, and technologies used by mature products such as ChatGPT. Since the technical solution of this application does not focus on this part, this application refers to this type of text generation big model.

[0042] Material library: It is a collection of a series of documents, which is a compilation of historical technical targets. Some technical target contents of the technical solution of this application are obtained from the highly similar contents in the material library.

[0043] Scoring criteria: A file submitted by the user, which contains scoring requirements for various aspects, indicating what the user values ​​in this technical specification. The technical solution of this application will guide the large model to generate the title and content of the technical specification based on the scoring criteria uploaded by the user.

[0044] Technical bids play a vital role in the bidding process. However, the process of manually writing technical bids from scratch is tedious and time-consuming, and requires users to have the ability to write technical bids proficiently. With the rapid development of large language models, combined with a large number of existing bid documents in the industry, technical bids can be effectively generated quickly according to user requirements.

[0045] The technical solution of this application can efficiently generate titles and corresponding text content related to user input information through a large text generation model and a material library. The large text generation model is responsible for generating various titles of technical targets and text content of some titles, and the remaining title content is completed by screening the content in the material library whose similarity with the generated title is higher than the set threshold. The combination of the large text generation model and the material library improves the quality and applicability of the text of the generated technical target.

[0046] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. The following description of at least one exemplary embodiment is actually only illustrative and is by no means intended to limit the present application and its application or use. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0047] In the following description, many specific details are set forth to facilitate a full understanding of the present application, but the present application may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present application. Therefore, the present application is not limited to the specific embodiments disclosed below.

[0048] Embodiment 1:

[0049] like Figure 1 As shown, the present invention discloses a device for generating a technical mark, including a data preparation unit, a title generation unit and a technical mark content generation unit, and each unit mainly works as follows.

[0050] The data preparation unit is used to collect and obtain the user's project profile, project theme and project scoring criteria. The project theme is obtained by giving the user optional options, taking the user's needs as the direction for generating the content of the bid, and generating a material library based on historical technical bid documents; the material library is a file that has been classified, titled, abstracted and technical bid content, and the material library is a docx or doc file; the content of each technical bid is marked with [start] as the beginning and [end] as the end, and each technical bid is distinguished by a number.

[0051] The following is part of the content of a material book classified as electronic intelligence.

[0052] Title: Project objectives, project status analysis

[0053] Abstract 0001: Construction objectives and needs analysis of scenic spot projects

[0054] Content 0001:

[0055] [start]

[0056] This plan focuses on the construction of XX smart scenic spot, with the construction of data center and command and dispatch center as the core, and the construction of various business application information systems as the link, to integrate scenic spot resources, realize information sharing, and innovate management models. Change decentralized management to collaborative linkage, change multi-level management to flat management, and change extensive management to spiritual management, to realize intelligent resource protection, intelligent business management, and information-based industrial integration, to use information to lead management, information to promote protection, and information to increase benefits, and comprehensively promote the sustainable development of the scenic area's environment, society, and economy. At the same time, the intelligent information resources of the scenic spot in various fields such as resource protection, planning and construction, tourism services, and standardized management are effectively integrated, and the scale effect is used to realize the maximum sharing of scenic spot customers, promote local economic development, and establish and improve a benign interactive mechanism between scenic spot resource protection and industrial development.

[0057] 1. Established a network framework system based on the supervision information system of 5A-level scenic spots;

[0058] 2. Increased the intensity of remote sensing monitoring and verification of 5A-level scenic spots;

[0059] 3. Improved the construction of the scenic area central database;

[0060] 4. Promoted the work of intelligent construction;

[0061] 5. Carry out network transmission of scenic area remote sensing monitoring data to provide reliable and efficient technical support for the supervision of 5A-level scenic areas.

[0062] The title generation unit is used to input the project overview and project theme as keywords into the text generation model according to the project overview, project theme and scoring criteria obtained by the data preparation unit, so as to generate a technical bid title; the text generation model reads the scoring criteria and project theme input by the user and the added industry experience or general requirements for titles, and scores the generated technical bid title; the technical bid title generation and title scoring are repeated 3 to 5 times, and the title corresponding to the highest score is selected as the final title.

[0063] For example, when using the API, the following prompts are used:

[0064] model="Your_model",

[0065] messages=[{"role":"user",

[0066] "content":f

[0067] You are a professional bid writer. According to the following scoring criteria and the theme of this project, generate a complete set of titles for the technical part, including first-level title, second-level title, third-level title, etc. The serial numbers must be accurate and as comprehensive as possible:\n\n Scoring criteria: {scoring_criteria}\n Project theme: {project_theme}\n Project overview {project_sum}, and the return value is json"}],

[0068] scoring_criteria is the scoring criteria entered by the user, project_theme is the project theme selected by the user, and project_sum is the project summary entered by the user.

[0069] The following example returns the json result, title is the title of the current node, count_title is the total number of titles of the current node, and children is a child node of the current title:

[0070]

[0071] An example of using the API to call text to generate a large model is as follows:

[0072] model="Your_model",

[0073] messages = [{

[0074] "role":"user",

[0075] "content":f"You are a professional tender reviewer, please evaluate in Chinese. Evaluate the following titles according to the given criteria:\n\nEvaluation criteria: {criteria}\nSchoosing criteria: {scoring_criteria}\nProject theme: {project_theme}\nTitle of the technical part: {title}"

[0076] }],

[0077] Criteria: The evaluation criteria are defined by the user based on the industry and needs, and do not require user input, such as:

[0078] criteria="

[0079] a. Read the title and rubric carefully.

[0080] b. Determine whether the title fully reflects the scoring criteria and project theme.

[0081] c. Evaluate the completeness of the title.

[0082] d. Assign a score from 1 to 5 to the title based on completeness

[0083] ”

[0084] scoring_criteria is the scoring criteria entered by the user.

[0085] project_theme is the project category entered by the user.

[0086] title is the generated title.

[0087] The scoring details are entered by the user on the interface. There will be a scoring details input box and the corresponding score. For details, see Table 1 for a sample of scoring details. The evaluation is performed by generating multiple batches of titles. The evaluation method combines the consistency of the technical part title with the scoring details and the project theme, the coverage of the title, the completeness of the title, and the consistency of the title to perform multi-dimensional scoring. The score is 1 to 5. The scoring function is completed by generating a large model with semantic understanding of text, such as GPT, Baichuan, Qianwen or ChatGLM.

[0088] For example, using a large model of text generation with semantic understanding to score the completeness of a title:

[0089] a. Read the title and rubric carefully.

[0090] b. Determine whether the title fully reflects the scoring criteria and project theme.

[0091] c. Evaluate the completeness of the title.

[0092] d. Assign a score from 1 to 5 to the title based on completeness.

[0093] Example results of evaluating the generated title are shown below.

[0094] The consistency between the title of the technical part and the scoring rules and project theme: 0

[0095] Title coverage: 0

[0096] Title completeness: 0

[0097] Title consistency: 5

[0098] Among them, after multiple scoring, the highest score can be taken after weighting according to various scoring standards and industry experience. The title corresponding to the highest score will be the final title.

[0099] A technical mark content generation unit is used to obtain multiple final titles according to the title generation unit, group the multiple final titles to obtain a material library title group and a large model title group, select from the material library according to each material library title in the material library title group to obtain the content corresponding to the material library title and obtain the content corresponding to the material library title group; according to each large model title in the large model title group, fill the project overview and the large model title as keywords into the prompt word, generate the content corresponding to the large model title through the text and obtain the content corresponding to the large model title group, merge the content corresponding to the material library title group and the content corresponding to the large model title group through the index during grouping, and generate the technical mark.

[0100] Embodiment 2:

[0101] like Figure 2 to Figure 5 As shown, the present invention discloses a method for generating a technical mark, which is used in the device for generating a technical mark described in Example 1, including the steps of S1 data preparation, S2 title generation and S3 technical mark content generation, which are described in detail as follows.

[0102] Step S1: Data preparation

[0103] Collect and obtain the user's project profile, project theme and project scoring criteria. The project theme is obtained by giving the user optional options. The user's needs are used as the direction for generating the content of the bid, and a material library is generated based on historical technical bid documents. The material library is a file with classification, title, summary and technical bid content. The material library is a docx or doc file. The content of each technical bid is marked with [start] as the beginning and [end] as the end, and each technical bid is distinguished by number.

[0104] like Figure 2 As shown, the data preparation unit in the technical mark generation device of Example 1 provides a user-friendly page to guide users to complete the upload of files. Users need to upload the project overview and project scoring details, and select the project type. The preparation of data is very important in the implementation of this device, which determines the direction of the title and content, so the strictness of data collection must be controlled. Therefore, in actual situations, the project category in data collection needs to be selected by the user to help users quickly locate. The project overview is supplementary content, and the importance weight will be appropriately adjusted when using the text generation model.

[0105] Step S1 specifically includes the following steps:

[0106] Step S101: The user inputs the project type, including but not limited to catering management, billboard installation, security and cleaning, etc. The data preparation unit provides detailed project types for the user to choose, aiming to generate technical specifications in a targeted manner;

[0107] Step S102: Enter the project overview, requiring the user to summarize the project in a short sentence;

[0108] Step S103: There are different scoring criteria for different project types. In order to reduce the user's operation threshold, the data preparation unit can select recommended scoring criteria. It also supports user customization needs and uploads scoring criteria compiled by users themselves to guide the generation of technical title and content. This application will guide the text generation model to generate titles and content based on the scoring criteria uploaded by users.

[0109] The scoring criteria is a document that contains multiple scoring points and the specific content for each scoring point. An example is shown in Table 1.

[0110] Table 1: Sample scoring rubric

[0111] Step S2: Title generation

[0112] According to the project overview, project theme and scoring criteria obtained by the data preparation unit, the project overview and project theme are input as keywords into the text generation model to generate a technical target title; the text generation model reads the scoring criteria and project theme input by the user as well as the added industry experience or general requirements for titles, and scores the generated technical target title; the technical target title generation and title scoring are repeated 3 to 5 times, and the title corresponding to the highest score is selected as the final title.

[0113] like Figure 3As shown, in the process of generating the technical mark by the title generation unit in the technical mark generation device of Example 1, the title of each chapter of the generated technical mark will also serve as an important basis for generating the text content of the technical mark, so the selection of the final title is crucial. In order to ensure that the generated title can meet the needs of users, this device will generate and evaluate the title multiple times, using different evaluation criteria, combining the evaluation criteria in the collected receipts and the industry evaluation criteria after the classification items, formulating rules through the guidance of relevant experts, and then letting the model evaluate the title according to these rules and score it, and finally selecting the title with the highest score as the generated title.

[0114] Step S2 specifically includes the following steps:

[0115] Step S201: extract the project overview, project theme and scoring details in the data preparation unit;

[0116] Step S202: The data retrieved by the data preparation unit in step S201 is sent to the keyword text generation model as the prompt word to generate a primary title and a secondary title; after the title is generated, all the generated titles are scored; the title with the highest score among the multiple batches of titles is selected as the title of the technical target; the details are as follows:

[0117] Step S202A: Generate a title for the technical mark using a text-based generation model based on the project overview, project theme and scoring criteria manually input by the user. The text-based generation model is generated by inputting the project overview and project theme as keywords of the text-based generation model.

[0118] Step S202B: The text generation model is used to read the scoring criteria and project theme entered by the user, as well as some of the user's own experience in the industry or some general requirements for titles, and score the title generated in step S202A.

[0119] Repeat the above steps S202A and S202B to generate titles and score 3 to 5 times, and select the title generated corresponding to the highest score as the final title.

[0120] After multiple scorings, the highest score can be taken based on weighted factors according to various scoring criteria and industry experience. The title corresponding to the highest score will be the final title.

[0121] Through step S202A and step S202B, the generated title is finally obtained.

[0122] Step S203: After obtaining a set of generated technical target titles according to step S202, we need to divide the titles into two groups. They are the material library title group and the large model title group. Material library title group: the content corresponding to the titles in this group will be selected from the material library. Large model title group: the content corresponding to the titles in this group will be automatically generated by the text generation large model.

[0123] Here are some grouping methods. The grouping itself is to balance the content of the selected material library and the text generation model. The proportion distribution problem needs to be determined based on the actual experience of the industry. If there is no proportion requirement and the number of material libraries is sufficient, a random proportion distribution method can be used.

[0124] like Figure 4 As shown, the material library group consists of multiple documents, each of which has multiple titles, summaries and contents.

[0125] Step S3: Generate technical specification content

[0126] According to the title generation unit, multiple final titles are obtained, and the multiple final titles are grouped to obtain a material library title group and a large model title group. According to each material library title in the material library title group, the content corresponding to the material library title is selected from the material library and the content corresponding to the material library title group is obtained; according to each large model title in the large model title group, the project overview and the large model title are filled into the prompt word as keywords, the content corresponding to the large model title is generated through text generation and the content corresponding to the large model title is obtained, and the content corresponding to the material library title group and the content corresponding to the large model title group are merged through the index during grouping to generate a technical label.

[0127] like Figure 5 As shown in the figure, the technical bid content is generated by using a large text generation model and a material library to generate technical bid content respectively, and then merge them in order according to the title. The large text generation model can provide personalized and innovative content, while the material library has higher credibility and professionalism. The content generated by the large text generation model can meet the personalized needs of users and inject novel elements and unique perspectives into the bid; while the content generated by matching the material library can draw on existing best practices and industry standards to ensure that the generated bid content has wide applicability and professionalism. The comprehensive use of the two methods can improve the diversity and quality of the generated content, reduce repetition, and reduce redundant information, so that the final generated bid is more refined and comprehensive.

[0128] Step S3 specifically includes the following steps:

[0129] Step S301: Based on the technical title generation, when the program determines the corresponding ratio of the material title group and the large model title group, the contents of the two title groups need to be supplemented respectively.

[0130] Step S301A: Supplement the content corresponding to the title group in the material library. What needs to be done is to select each material that meets the title content from the material library. The selection method is as follows:

[0131] Step S301Aa: Sort the similarity between one of the generated titles and the abstracts of the materials in the material library, exclude the selected content through the program, and then select the remaining content with the highest similarity as the content of the title generated based on the material library.

[0132] Step S301Ab: Repeat step S301Aa for all titles in the material library title group. Obtain a content group corresponding to the title generated by the material library - the material library content group. The similarity can be calculated by cosine similarity, Euclidean distance, word vector similarity based on a pre-trained model, or a text matching model, and the text matching model is RNN, LSTM, GRU or Transformer.

[0133] The following data is based on the similarity calculation of word vectors. The first array in the array is the similarity matrix between the summary of the material library and one of the generated titles, and the scores array is their similarity.

[0134] "text_embedding":array([[-0.22365333,0.4202284,0.2666558,...,-0.26484376,

[0135] 0.40744498,0.2772795],

[0136] [-0.25315338,0.38203242,0.24046001,...,-0.3280005,

[0137] 0.41472986,0.29768166],

[0138] [-0.24323441,0.41074485,0.24910843,...,-0.30696347,

[0139] 0.40286088,0.27363694],

[0140] [-0.25041893,0.3749906,0.24194777,...,-0.3197233,

[0141] 0.41340476,0.27778062]],dtype=float32),'scores":[70.5133285522461,70.56582641601562,70.45124816894531]}

[0142] Step S301B: Supplement the content corresponding to the large model title group. The content of the large model title group needs to be generated through the text generation large model. The collected project profile and the generated title need to be filled into the prompt word as keywords to obtain a large model content group based on the text generation large model to obtain the generated title corresponding content.

[0143] Take the API for calling text to generate a large model as an example:

[0144] messages=[

[0145] {"role":"system","content":"You are a skilled bid document writer. You write project proposals based on the elements provided by users."},

[0146] {"role":"user","content":f"Please go directly to the main text and write specific content about {title} based on the information provided by {project_sum}."} ]

[0148] Step S302: Finally, the material library content group and the large model content group are merged through the index during grouping to generate a complete technical label.

[0149] Embodiment 3:

[0150] The present invention discloses a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program including a data preparation unit, a title generation unit, and a technical mark content generation unit of embodiment 1, and when the computer program is executed by a processor, the steps in embodiment 2 are implemented.

Claims

1. A device for generating a technical mark, characterized in that: It includes a data preparation unit, a title generation unit and a technical bid content generation unit. The data preparation unit is used to collect the user's project profile, project theme and project scoring criteria, and generate a material library based on historical technical bid documents; The title generation unit is used to input the project overview and project theme as keywords into the text generation model according to the project overview, project theme and scoring criteria obtained by the data preparation unit, so as to generate a technical bid title; the text generation model reads the scoring criteria and project theme input by the user and the added industry experience or general requirements for the title, and scores the title of the generated technical bid; repeatedly generates the technical bid title and scores the title, and selects the title corresponding to the highest score as the final title; A technical mark content generation unit is used to obtain multiple final titles according to the title generation unit, group the multiple final titles to obtain a material library title group and a large model title group, select from the material library according to each material library title in the material library title group to obtain the content corresponding to the material library title and obtain the content corresponding to the material library title group; according to each large model title in the large model title group, fill the project overview and the large model title as keywords into the prompt word, generate the content corresponding to the large model title through the text and obtain the content corresponding to the large model title group, merge the content corresponding to the material library title group and the content corresponding to the large model title group through the index during grouping, and generate the technical mark.

2. The device for generating a technical mark according to claim 1, characterized in that: The data preparation unit is also used to obtain the project theme by giving users optional options, and taking the user's needs as the direction for generating the content of the tender. The material library is a file with classification, title, summary and technical specification content, and the material library is a docx or doc file; the content of each technical specification is marked with [start] as the beginning and [end] as the end, and each technical specification is distinguished by a number.

3. The device for generating a technical mark according to claim 1, characterized in that: The title generation unit is also used to repeatedly generate the title of the technical mark and the title score 3 to 5 times.

4. A method for generating a technical mark, used in a device for generating a technical mark as claimed in any one of claims 1 to 3, characterized in that: It includes the steps of S1 data preparation, S2 title generation and S3 technical bid content generation. S1 data preparation collects the user's project profile, project theme and project scoring criteria, and generates a material library based on historical technical bid documents; S2 title generation, based on the project profile, project theme and scoring criteria obtained by the data preparation unit, inputs the project profile and project theme as keywords into the text generation model to generate the technical bid title; the text generation model reads the scoring criteria and project theme input by the user and the added industry experience or general requirements for the title, and scores the title of the generated technical bid; repeatedly generates the technical bid title and title scoring, and selects the one with the highest score. The title is taken as the final title; S3 technical mark content is generated, multiple final titles are obtained according to the title generation unit, the multiple final titles are grouped to obtain a material library title group and a large model title group, and according to each material library title in the material library title group, the content corresponding to the material library title is selected from the material library and the content corresponding to the material library title group is obtained; according to each large model title in the large model title group, the project overview and the large model title are filled into the prompt word as keywords, the large model is generated through text to generate the content corresponding to the large model title and the content corresponding to the large model title group is obtained, the content corresponding to the material library title group and the content corresponding to the large model title group are merged through the index during grouping to generate the technical mark.

Citation Information

Patent Citations

  • Method and device for generating article based on artificial intelligence

    CN106874248A

  • Article title optimization method and system, medium and equipment

    CN110795930A

  • Content generation method, electronic equipment and storage medium

    CN117609483A

  • Article automatic generation method and device, equipment and storage medium

    CN117744620A

  • Item classification method and selection system for electronic solicitation

    US20160335674A1